Research on Autoregressive Ship-radiated Noise Synthesis Technology Based on Prompt Signal Constraints

Ship-radiated noise signal synthesis can provide important support for underwater acoustic identification algorithm research, acoustic simulation deduction, underwater acoustic deception, etc. Traditional ship-radiated noise synthesis technology is based on physical statistical models, and simplifies modeling through hypothetical conditions. The synthesized signals have certain limitations and insufficient authenticity. This article uses Deep Learning methods to model, and builds a synthesis system through the cascade of a time-domain signal encoder-decoder and a Transformer decoder. Under the constraint mechanism of the prompt signal, the ship-radiated noise is autoregressively synthesized. Experimental results show that the synthesized signal based on a few real signal prompts has a high similarity to the real signal.